摘要
针对高速列车受电弓在运行中与接触网接触不良产生拉弧,拉弧产生的环境复杂在线检测困难的问题。该文将图像处理与深度学习目标检测相结合,设计了一种基于改进SSD的受电弓拉弧目标检测模型。通过对模型的训练数据集模糊、加噪等处理来扩充数据集,模拟受电弓运行的各种复杂环境,提高模型的鲁棒性。其次,SSD模型基础网络用轻量型MobileNet网络,减小了模型的计算量,提高检测速度。并对采集的待检测弓网图像进行预处理,增强了弓网图像的特征信息,进一步提高了受电弓拉弧的平均检测精度。实验得出该方法能够满足列车受电弓拉弧检测需求,检测性能好,便于加载到列车上使用。
In order to solve the problem that the pantograph of high-speed train has poor contact with catenary in operation,the arc drawing environment is complex and the online detection is difficult.This paper combines image processing with deep learning target detection,and designs a detection model of pantograph arcing target based on improved SSD.The training data set of the model is fuzzy and noisy to expand the data set,simulate the complex environment of pantograph operation,and improve the robustness of the model.Secondly,the basic network of SSD model uses the light-weight mobile net network,which reduces the calculation of the model and improves the detection speed.Moreover,preprocessing the collected pantograph catenary image to enhance the characteristic information of pantograph catenary image,and further improve the average detection accuracy of pantograph arc.The experimental results show that the method can meet the demand of pantograph arc detection,and has good detection performance,which is easy to load on the train.
作者
冯庆胜
牛学慧
FENG Qing-sheng;NIU Xue-hui(School of Electrical Information Engineering,Dalian Jiaotong University,Dalian 116028,China)
出处
《自动化与仪表》
2021年第3期45-49,共5页
Automation & Instrumentation
基金
辽宁省自然科学基金资助项目(2019-ZD-0094)。
关键词
高速列车
改进SSD
目标检测
图像预处理
受电弓拉弧
high speed train
improved single shot multibox detector
target detection
image preprocessing
pantograph arcing
